Repository navigation
Releases: RockeyRoc/DeepProf
Release list
DeepProf v0.6.4 — browser workspace and BKT/RAG evidence
DeepProf v0.6.4 brings the local CLI/Gateway and browser chat/study workspace together with a curated BKT/RAG experiment export.
- Browser sessions support chat and explicitly labelled non-experiment study, document/image uploads, streamed replies, cancellation, model/reasoning controls and source inspection.
- Provider-aware native web search and deep-research reports distinguish a successful request from verified returned sources.
- Release packaging reads the package version, validates the checked-out tag, and publishes SHA-256 manifests alongside the CLI archive and public experiment bundle.
The latest full-text RAG batch contains 480 valid cells across 120 questions and four conditions, with 13,710 retrieval, 1,595 citation and 120 answerability AI labels. NoMIRACL Chinese evaluates 3,770 queries and 37,599 fixed-candidate pairs (test FAR 4.51%, FRR 70.11%, AUC 0.8007). These measurements do not establish student learning gain; genuine personnel review remains pending. BKT research candidates remain exploratory and do not replace the runtime development defaults. Historical failures and stopped batches remain preserved.
Assets: deepprof-cli-0.6.4.tgz, its checksum, a release-wide SHA-256 manifest, the Chinese Markdown/PDF report, a public evidence ZIP and its file manifest. Raw prompts, textbook pages, private databases and credentials are excluded. Node.js 22+ and Python 3.12+ are required; first setup needs Python package-index access. Existing user data and provider configuration are retained.
npm install --global https://github.com/RockeyRoc/DeepProf/releases/download/v0.6.4/deepprof-cli-0.6.4.tgz
deepprof doctor
deepprof web中文版本、接口与实验边界详见仓库 docs/RELEASEv0.6.4.md。人员复核与真实学生学习效果仍未完成；本版发布的是工程功能和可核对的实验记录。
DeepProf v0.6.2 — M1–M3 experiment report
This release combines the DeepProf v0.6.2 local CLI/Gateway, course-scoped retrieval and replay, versioned attempt/BKT path, and the M1–M3 experiment record.
The M1–M3 fixtures and evaluations are engineering evidence, not student learning outcomes. M3 live trial output had 26 length truncations among 29 provider requests; no group-effect conclusion is supported. Teacher review, parameter calibration and learner-outcome evaluation remain open.
Assets include the local-install CLI npm archive, the Chinese Markdown/PDF experiment report, and a redacted reproducibility bundle with source data, figures, scripts and SHA-256 manifests. Installation requires Node.js 22+, Python 3.12+ and network access to the Python package index on first setup.
DeepProf v0.6.1 — MVP Preview
English
DeepProf v0.6.1 is an MVP-1–MVP-5 engineering preview.
Included
- DeepProf Runtime with sessions, events, providers, memory, tools, storage, plugins, and sandbox boundaries.
- Pedagogical Graph with auditable PedagogicalDecision output and Runtime execution.
- Electron Desktop workbench, keyboard-first CLI, shared Session/Event contracts, resource library, and Pet package.
- Bilingual README and acceptance records for MVP-1 through MVP-5.
Verification
- Python full test suite passed.
- CLI typecheck, build, and 3 Node tests passed.
- Desktop typecheck and production build passed.
- Pet package validation passed.
Known limitations
BKT/IRT estimators, remaining streaming fallback wiring, plugin composition-root installation, live-provider quota/model availability, and signed installer packaging remain follow-up work. This release is intentionally marked pre-release.
中文
DeepProf v0.6.1 是 MVP-1～MVP-5 的工程预览版。
包含内容
- DeepProf Runtime：Session、事件、Provider、Memory、Tool、Storage、Plugin 与 Sandbox 边界。
- 可审计的 Pedagogical Graph：输出 PedagogicalDecision，由 Runtime 负责执行。
- Electron Desktop 工作台、键盘优先 CLI、共享 Session/Event 契约、资源库和桌宠包。
- 中英文 README 以及 MVP-1～MVP-5 验收记录。
验证结果
- Python 全量测试通过。
- CLI typecheck、build 和 3 项 Node 测试通过。
- Desktop typecheck 与生产构建通过。
- 桌宠包校验通过。
已知限制
BKT/IRT 估计器、流式 fallback 剩余接线、插件组合根安装、真实 Provider 配额/模型可用性和签名安装程序仍需后续完善。本版本明确标记为预发布版。
桌宠模块 v0.2.3
桌宠前端模块 v0.2.3。
附件名为纯英文是故意的 —— GitHub 的附件名参数对中文处理不可靠。
源码分支 feat/desktop-module（tag 打在这个分支上，main 未动）。
桌宠模块 v0.2.2
桌宠前端模块 v0.2.2。
附件名为纯英文是故意的 —— GitHub 的附件名参数对中文处理不可靠。
源码分支 feat/desktop-module（tag 打在这个分支上，main 未动）。
桌宠模块 v0.2.1
桌宠前端模块 v0.2.1。
附件名为纯英文是故意的 —— GitHub 的附件名参数对中文处理不可靠。
源码分支 feat/desktop-module（tag 打在这个分支上，main 未动）。
桌宠模块 v0.2.0（官方 Q 版角色 + 走路 + 眨眼）
桌宠前端模块 v0.2.0 —— 换官方 Q 版角色后的第一次可运行交付。
⚠️ 这一版不是 0.1.0 的延续：0.1.0 用的是上一个角色（深蓝双马尾），
双击会看到旧形象。请用本版。
这一版有什么
- 角色换成官方 Q 版（白发 + 蓝蝴蝶结 + 水手风），6 张表情立绘
- 走路解禁 —— 侧视四格图切成 4 帧
- 眨眼解禁 —— 站姿睁眼/闭眼两帧，每 3~7 秒随机眨一次
两个附件怎么选
| 附件 | 是什么 |
|---|---|
DeepProf-Pet-portable-0.2.0.exe |
免安装版（portable）—— 演示首选，双击即跑，不用装 Node |
DeepProf-Pet-setup-0.2.0.exe |
安装版（NSIS），可选安装目录、建桌面快捷方式 |
⚠️ 附件名是纯英文，这是故意的：GitHub 的附件名参数对中文处理不可靠，
两个中文文件名会被压成同一个乱码名传不上去。程序本身界面是中文的，只有文件名是英文。
首次运行会有 Windows SmartScreen「未知发布者」警告 —— 没有代码签名，点「仍要运行」即可。
也没有应用图标，用的是 Electron 默认图标。两条都是已知遗留。
还差什么（需要美术再出图）
- 同姿势的闭嘴图 × 1 张 —— 现在给的两张都是张嘴的，做不出「说话嘴部开合」
- 待机 2 帧 —— 呼吸起伏
- hint / correct / quiz 各 1 张 —— 这 3 个教学槽位目前在复用 think/teach
源码
分支 feat/desktop-module（tag 打在这个分支的 HEAD 上，main 未动）。
本次改动一页纸汇总：desktop/docs/本次改动汇总_20260920.md